VessQC

A napari plugin for uncertainty-based quality control and manual correction of 3D blood vessel segmentations.

---

Overview

VessQC is a plugin for the image viewer napari. It is designed for the inspection and correction of 3D vessel segmentations generated by machine-learning pipelines.

The plugin combines:

• visualization of image, segmentation, and uncertainty data,
• automatic grouping of uncertain regions into segments,
• interactive review of suspicious regions,
• manual correction of segmentation masks,
• export of corrected segmentation results.

The plugin is particularly useful for workflows in which:

• a neural network generates vessel segmentations,
• uncertainty maps are available,
• only uncertain regions should be reviewed manually.

---

Screenshot

---

Installation

Install the released version from PyPI:

pip install VessQC

Install the latest development version from GitHub:

pip install git+https://github.com/MMV-Lab/VessQC.git

---

Input Data

VessQC expects three volumetric datasets:

1. Original image data
2. Predicted vessel segmentation (*_segPred)
3. Uncertainty map (*_uncertainty)

Supported file formats:

• TIFF (.tif, .tiff)
• NIfTI (.nii, .nii.gz)

Example:

Sample_IM.tif
Sample_segPred.tif
Sample_uncertainty.tif

---

Usage

1. Load the image

Press:

Load image

and select the original 3D image.

---

2. Load segmentation and uncertainty data

Press:

Read segPred file

The plugin automatically searches for matching:

• *_segPred
• *_uncertainty

files in the same directory.

The uncertainty map is segmented into connected uncertain regions.

---

3. Review uncertain segments

Press:

Show list of segments

A separate window displays all detected uncertain regions.

For each segment the following information is shown:

• segment name,
• uncertainty value,
• voxel count,
• processing state.

---

4. Inspect a segment

Click a segment button to:

• zoom into the corresponding region,
• crop the surrounding image data,
• display the local segmentation,
• center the viewer on the selected segment.

---

5. Correct the segmentation

Use the standard napari label editing tools to:

• add missing vessel voxels,
• remove incorrect voxels,
• refine the segmentation mask.

After correction press:

done

The modifications are transferred back into:

• the segmentation volume,
• the uncertainty map,
• the internal label representation.

---

6. Save the result

Press:

Save final result

The corrected segmentation is written to disk as:

*_segPred_New.tif

Optionally the corrected uncertainty map can also be saved.

---

Intermediate Data

VessQC can store intermediate processing results in the temporary directory.

The following data are saved:

• segmentation volume,
• uncertainty volume,
• label volume,
• segment metadata.

This allows interrupted curation sessions to be resumed later.

---

Development

Run the test suite with:

pytest

---

Contributing

Contributions and bug reports are welcome.

Please include:

• a detailed problem description,
• steps to reproduce the issue,
• example data if possible.

---

License

Distributed under the terms of the BSD-3 license.

---

Repository

GitHub repository:

https://github.com/MMV-Lab/VessQC